use get_device_capability since CI setting in cfg is unreliable

src/axolotl/utils/models.py

@@ -633,7 +633,7 @@ class ModelLoader:
             if self.cfg.use_flash_attention_3 is True:
                 use_fa3 = True
             elif self.cfg.use_flash_attention_3 == "auto":
-                if int(self.cfg.capabilities.compute_capability.split("_")[-1]) >= 90:
+                if torch.cuda.get_device_capability() >= (9, 0):
                     # FA3 is only available on Hopper GPUs and newer
                     use_fa3 = True
             if use_fa3 and importlib.util.find_spec("flash_attn_interface") is not None: